Object Detection Model Based On GAN Region Of Interest

Object detection is one of the important research directions in the field of computer vision.It integrates positioning,identification and classification tasks.It is widely used in many areas such as intelligent monitoring and human interaction.Among them,the object detection model uses the deep neural network to extract the target features,and the region proposal network is used to synchronize the extracted target features and the classification tasks,which significantly improves the detection accuracy.In deep neural networks,the Generative Adversarial Network is a unique network that captures the distribution of potential data.The design of true and false counter loss based on game theory makes it possible to cope with different types of tasks.In recent years,it was widely used in the fields of image translation,image generation,and natural language.On this basis,this paper conducts an in-depth study of the mainstream deep learning object detection model and the Generative Adversarial Networks.This paper first introduces the research background and significance of the object detection model and the Generative Adversarial Networks,then analyzes the mainstream algorithm principles of the deep learning object detection model,such as RCNN,FastRCNN and SSD,and the bottleneck problems in these algorithms.The main work and innovations of this paper are:(1)This paper studies the traditional Generative Adversarial image mapping model,and the shortcomings of the Generative Adversarial image mapping model in the three-channels to the single-channel image.We achieve a foreground-background separation model to realize the mapping translation of real image target binarized image.(2)This paper studies the current mainstream deep learning object detection algorithm.The detection accuracy of SSD-based object detection model is limited to the size of datasets.This paper proposes a data enhancement based on the foreground-background separation model.Then we combine the original object detection algorithm with the foreground-background separation model.By not increasing the complexity of the original algorithm model,the object detection accuracy is improved by this training method.The improved object detection model was trained under the PASCAL VOC dataset.As an input,the test was achieved at the image scale of 321×321,which achieved 78.7%mAP in the VOC2007-test dataset and 76.6%mAP in the VOC2012-test dataset.